kayakstream processing you can watch running

describe pipelines as input → transforms → output, and kayak runs them on a live canvas — a card per pipeline, edges for the data, a log and a throughput chart on each one.

getting started

component reference

github

configured, not coded

nats, kafka, mqtt, redis, opc ua and http in; postgres, clickhouse, s3, files and webhooks out; filter, reduce, map, split and remember in between. plain JSON the whole way through — no schema to define up front.

a graph, not a list of jobs

a pipeline can feed another pipeline, so what you configure is a DAG. the canvas is a real view onto the running server: watch it, edit it, or drive the same JSON API by hand.

the reference cannot rot

every table under reference/ is reflected out of the config types kayak actually deserializes, and the HTTP tables come from the table the router is built from. no component is documented by hand.